1. science
2. mathematics
3. statistics

How-to in R
Wiebe R. Pestman
Contents
1
Histograms
1
2
Bar plots and pin diagrams
2
3
Cumulative frequency plots
3
4
Saving graphics
3
5
Numbers, strings and objects in R
4
6
Writing data to plain text files
9
7
Reading in data that is stored in plain text files
9
8
Saving data to native file format
11
9
Loading data that is stored in native file format
11
10 Loading built-in data into your workspace
11
11 Importing foreign data
12
12 Importing Excel data
13
13 Reading in tabulated data
13
14 Generating random data
14
15 Mean and median
14
16 Variance
14
17 Transforming data
15
18 Quantiles and boxplots
15
19 Making your own functions
15
20 Saving functions
16
21 Plotting curves
17
22 Plotting straight lines
18
23 Kolmogorov-Smirnov plots
18
24 Kolmogorov-Smirnov distance
19
25 QQ-plots
20
26 Probability distributions
21
27 Plotting probability densities
22
28 Confidence intervals for a population mean
23
29 The 1-sample t-test
23
30 The 2-sample t-test
24
31 The paired-sample t-test
24
32 Kolmogorov-Smirnov tests
24
33 Fisher’s variance test
25
34 Splitting datasets
26
35 The 1-way ANOVA
26
36 The 2-way ANOVA
27
37 Wilcoxon’s signed-rank test
28
38 The Mann-Whitney test
29
39 The Kruskal-Wallis test
29
40 The runs test
29
41 Simple linear regression
30
42 Multiple linear regression
31
43 Regression plots
31
44 Generating random regression data
32
45 Binomial tests
32
46 Proportion tests
33
47 Contingency tables
33
48 McNemar’s test
33
49 The chi-square test on goodness of fit
34
50 The chi-square test on independence
35
51 Fitting nonlinear models
36
52 Residual analysis
37
53 Logistic regression
38
54 Nonparametric survival analysis
39
55 Power and sample size (I)
39
56 Power and sample size (II)
40
A Installation of R-base and R-libraries
42
B How to make your own libraries under R?
B.1 Building an R-library under Windows . . . . . . . . . . . . . . . .
B.1.1 Installation of compilers . . . . . . . . . . . . . . . . . . .
B.1.2 Compiling a binary installer for an R-library under Windows
B.1.3 How to use the newborn installer . . . . . . . . . . . . . . .
B.2 Building an R-library under Unix-systems . . . . . . . . . . . . . .
B.3 Building an R-library under MacOSX . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
44
44
44
44
45
45
46
1
Histograms
How to make a histogram in R? In this section you will learn how to do this. To have some data
to work with, create a sequence height of 50 numbers by passing the following command (this
command is be explained in the section Generating random data):
height <- rnorm(50,175,10)
and strike the key Enter. It will seem as if nothing has happened. Pass the command
height
and strike the key Enter. You will then see a list of 50 numbers indeed: You have been storing
the sequence of 50 numbers in an object height that resides in the so-called workspace of R (see
the section Numbers, strings and objects in R). Pass the command
ls()
or
objects()
to see a complete list of all the objects in your workspace.
To create a histogram of the sequence height, just pass the command
hist(height)
and strike the key Enter. A graphics window will then appear with in it a histogram presentation
of the sequence height.
To see the extensive possibilities of the histogram command, just type
?hist
Play around a bit with this command. For example, change the colour of the histogram and the
number of intervals. Use the option xlab="blablabla" to add a label to the horizontal axis.
By using ylab istead of xlab you may add labels to the vertical axis. See the section Plotting
curves to learn how a normal curve can be displayed in the histogram.
If you want to remove the object height from the workspace, enter the command
rm(height)
The object height is then removed from your workspace. To check this, just pass the command
ls()
as you did earlier.
1
2
Bar plots and pin diagrams
How to make a bar plot in R? As an example, let us take the table below:
score
1
2
3
4
5
6
frequency
71
113
83
134
103
96
Presenting this table graphically as a bar plot in R can be done as follows: First put in an object,
say freq, like this
freq <- c(71,113,83,134,103,96)
Then pass the command
barplot(freq)
and your bar plot will appear. In this bar plot, however, the scores are not indicated on the horizontal axis. To bring this about, do the following. First make an object, say scores that contains
the scores.
spots <- c(1,2,3,4,5,6)
Then weight the scores with the frequencies stored in the object freq like this
john <- rep(spots,freq)
Print the object john to the screen to see its contents. Run the command
table(john)
Finally, run the command
barplot(table(john))
This will provide the desired bar plot.
The table can also be presented as a pin diagram. To this end, pass the command
plot(spots,freq)
You will see a figure appear, but it is not the type of figure you wish to see. Close the figure and
repeat the above with the option to specify a type
plot(spots,freq,type="h")
You will then get your desired pin diagram. Just for fun, also try
plot(spots,freq,type="l")
Here "l" stands for the 12th character in the alphabet; it is not the number 1. The graphic utility
plot() in R is very versatile. You can do a lot of graphic work with it. Type
?plot
to get an impression of the possibilities of this utility.
2
3
Cumulative frequency plots
How to make a cumulative frequency plot in R? To illustrate this, produce some data to work with
and store it in the object x (for example the flood size data of the Ocmulgee River)
x <- c(50,12,16,20,17,13,61,26,33,38)
Next, sort the values in x from low to high
x <- sort(x)
Now store the length of the sequence x in an object n:
n <- length(x)
So, in this example, n has the value 10. Define the cumulative frequencies as
y <- (1:n)/n
Print the values of y to the screen to see what you have been doing by the above.
Next, the graphic utility plot() can be used to produce the cumulative frequency plot:
plot(x,y,type="s")
If you want to include labels on the x-axis and y-axis, then do something like
plot(x,y,type="s",xlab="Size of Flood")
Similarly a label can be attached to the y-axis. Don’t forget the quotation marks!
Alternatively, probably easier, a cumulative frequency plot can be produced by means of the function ecdf(). It is not necessary to sort the sequence x then. Just run the command
john <- ecdf(x)
By this action the empirical distribution function is created and named john. If you type the
command john(27) then you get the value 0.6 returned. This indicates that 60% of the data in
x is less than 27. A plot can be obtained by passing a command of type
plot(john,verticals=TRUE,do.points=FALSE)
Also try this without the options vericals=TRUE and do.points=FALSE.
4
Saving graphics
The graphics produced in an R-session can of course be saved to files. The function dev.copy(),
does the job in this. You can save a given plot on your screen as follows:
dev.copy(jpeg,file="X:/path/to/mygraph.jpg")
Here the letter X stands for the drive on which you want to have your file saved. By the above
a file mygraph.jpg is created and a bitstream towards this file is started. To close the stream,
pass the command
dev.off()
3
Your figure is now saved in a jpg file X:/path/to/mygraph.jpg. If you simply type
dev.copy(jpeg,file="mygraph.jpg")
dev.off()
then the file mygraph.jpg is saved in the working directory. Several options are available to
tune the process. For example, one can set the width and height of the picture (in pixels) as follows:
dev.copy(jpeg,file="X:/path/to/mygraph.jpg",width=200,height=250)
dev.off()
Don’t forget to close the bitstream by passing the command dev.off(). If you don’t, it could
result in misshapen files. Besides width and height, one could also set the background color and
the orientation of the figure as follows:
dev.copy(jpeg,file="/path/to/mygraph.jpg",width=200,height=250,
bg="red",horizontal=F)
dev.off()
In a similar way one can save graphics in the formats png, pdf and eps. In eps the command would
be something like:
dev.copy(postscript,file="mygraph.eps",width=5,height=5)
dev.off()
Rather than in pixels, in postscript (eps) the width and height of the figure must be put in in inches!
Note that, when running R on a Windows machine, you can also use the function savePlot()
to save your graphics. Pass the command ?savePlot to get information about this method.
5
Numbers, strings and objects in R
When working in R it is comfortable to know a little bit about bits, bytes, strings, objects etcetera.
This section is a quick guide to these kind of things. The explanation will be based on similarity
to things that you have learnt at school during chemistry lessons. In one of your first chemistry
lessons you probably were taught that atoms are thought to be the building blocks in this science.
By means of these atoms one can build molecules. Molecules, in turn can be ranked in molecular
arrays, such as lattices, chains, double helices etcetera.
In computer sciences the atoms can be thought to be characters and digits. Examples of characters
are the symbols "a", "B", "!", "y", "@", ";" etcetera. A space, that is " ", is a less trivial example
of a character. Digits are understood to be the symbols 0, 1, 2, . . . , 8, 9. Digits distinguish
themselves from characters in that they allow for operations like multiplication, addition and so
on (see footnote). Characters fail to have these properties. Thus the atoms of informatics come in
two flavors: characters and digits. Note that in chemistry atoms also come in flavors: metals, inert
gases, halogens, . . .
The molecules in informatics are numbers and strings. One can bind together digits to form
numbers, for example 112, 1001, 47, 0.37, . . . . Characters can be compounded into so-called
A highly confusing thing is here that the symbols 0, 1, 2, . . . , 9 can also be used as being characters, that is, as characters
"0", "1", "2", . . . , "8", "9". They just serve as symbols then; so one cannot multiply the character "3" by the character
"2".
4
strings, for example "Avenue Saint Laurent; Montreal". Note that the latter molecule (or string)
contains spaces among its atoms.
In chemistry the molecules often build chains, lattices, cycles etcetera. In informatics, specifically
in R, the molecules can be used (or stored) in so-called objects. Rather than chains, crystals,
cycles, etcetera, the objects in R are called vectors, matrices, data frames, lists and factors. Below
you can find a brief description of these types of objects.
vectors A vector in R could be thought to be the equivalent of a chain of molecules in chemistry.
Vectors are sequences of numbers and/or strings. Vectorial objects can be regarded as being
the most simple objects that R offers the user. To give an example, with the three molecules
or strings "john", "peter" and "nicholas" one could build a vector (chain, sequence) in R,
named myfirstvector, like this
myfirstvector <- c("john","peter","nicholas")
By this action R creates an object, named myfirstvector, that contains the strings
"john", "peter" and "nicholas" as its basic components. The components of the vector
myfirstvector can be displayed on the screen by simply typing myfirstvector
and striking the key Enter. Order matters! The vector defined by
mysecondvector <- c("peter","nicholas","john")
differs from myfirstvector! A vectorial object may also contain just numbers; try the
following in R:
mythirdvector <- c(47,21,10,9)
A component, say the 4th component of the vector mythirdvector can be accessed by
typing
mythirdvector[4]
Components can be overwritten like this:
mythirdvector[4] <- 11
mythirdvector[2] <- "jane"
The 3th component can be deleted from the vector by typing
mythirdvector[-3]
If the component’s of a vector are all numerical, then one may apply arithmetic operations
to it. For example, one may evaluate
3*mythirdvector
Last but not least, a vectorial object in R may contain both numbers and strings:
myfourthvector <- c(23,"john",10,9)
5
Arithmetical operations are forbidden for vectors hat contain strings among their components.
matrices A matrix in R could be thought to be the equivalent to a 2-dimensional (or flat) rectangular lattice in chemistry. A matrix has columns and rows. To see an example in R, first
make a vector with 6 components:
myvector <- c(2,1,9,3,3,4)
Now turn this vector into a matrix with 3 rows and 2 columns, that is a 3x2-matrix, as
follows:
myfirstmatrix <- matrix(myvector,ncol=2)
Print the contents of myfirstmatrix to the screen to see what you created through the
above. The matrix myfirstmatrix could also have been created via the command
myfirstmatrix <- matrix(myvector,nrow=3)
The element on row number 3 and column number 2 can be accessed like this
myfirstmatrix[3,2]
Row number 3 and column number 2 can be accessed as follows:
myfirstmatrix[3,]
myfirstmatrix[,2]
Row number 3 can be deleted from the matrix by running
myfirstmatrix[-3,]
In a similar way a column can be deleted. To transpose a matrix, that is to say, to turn the
rows into columns (and the columns into rows) do the following:
mysecondmatrix <- t(myfirstmatrix)
The elements of a matrix are either numbers or strings; it is not possible to store both strings
and numbers in one and the same matrix. Numerical matrices can be subjected to multiplication and addition, as defined in linear algebra. For example, the matrices myfirstmatrix
and mysecondmatrix can be multiplied through the command
mythirdmatrix <- myfirstmatrix %*% mysecondmatrix
Thus the power of linear algebra can be exploited.
The rows of the matrix myfirstmatrix can be assigned names as follows
rownames(myfirstmatrix) <- c("john","peter","nicholas")
The columns can be equipped with headers like this:
6
colnamesnames(myfirstmatrix) <- c("mary","jane")
Print the contents of myfirstmatrix to the screen to see the effect of the above.
arrays An array is a generalized matrix. Generalized in the sense that the matrix is allowed to be
of multi-dimensional size. A 2-dimensional array is just a matrix. A 3-dimensional array
could be compared to a 3-dimensional cubic lattice of molecules. Alternatively it could be
compared to a book the pages of which consist of matrices. As an example, given any vector
john of length 120, it can be turned into a 3-dimensional array peter of size 4x5x6 like
this
peter <- array(john,dim=c(4,5,6))
Now the object peter can be regarded as a book with 6 pages. Each page consists of a
4x5-matrix. The entry on row 3 and column 2 of the matrix on page 5 can be accessed by
typing
peter[2,3,5]
Page 3 of the book peter can be accessed as
peter[,,3]
The same array could, of course, also be considered as a book with 3 pages where each page
consists of a 5x6-matrix. Page 2 could then be accessed as
peter[2,,]
Arrays of higher dimensions can also be created. For example, one could create a 4dimensional array nicholas of size 2x2x5x6 as follows:
nicholas <- array(john,dim=c(2,2,5,6))
The object nicholas could be compared to a library that contains 6 books. Each book is
an array of size (2, 2, 5).
lists In a list you can store virtually all kinds of things. To illustrate this, create three objects in R
like this
x <- c("john","peter","nicholas")
y <- c("mary","jane")
z <- matrix(1:12,ncol=3)
These three objects can be stored in a list in the following way:
myfirstlist <- list(x,y,z)
Now the object myfirstlist contains the objects x, y and z as elements. You can access
an object, say the object y, from the list by typing
myfirstlist[[2]]
7
The components of the list can be equipped with headers, say the headers garcons,
filles and mat, by redefining it as
myfirstlist <- list(boys=x,girls=y,matrix=z)
The three components of the list can then also be accessed by typing
myfirstlist$boys
myfirstlist$girls
myfirstlist$matrix
A list may even contain lists as components:
mysecondlist <- list("hillary",myfirstlist)
In this list the object x can be accessed by typing
mysecondlist[[2]][[1]]
or, by using the headers in myfirstlist, by typing
mysecondlist[[2]]$boys
data frames Data frames resemble matrices in that they have columns and rows, but under the
hood they are lists. They can be accessed as if they were matrices. Contrary to a matrix,
however, a data frame may contain both numbers and strings. The price to be paid is that
one cannot apply linear algebraic operations to data frames. For example, two data frames
cannot be multiplied. Data frames are very popular storage objects among R users. This
type of objects approaches the concept of a spreadsheet.
factors Factors resemble vectors, but under the hood they differ from vectors. If the purpose of
the components of a vector is to define categories, that is, if the vector is used as a grouping
variable, then it may sometimes be necessary to convert a vector into a factor. This is, for
example, the case if you want to use a vectorial grouping variable as an explanatory variable
in a regression analysis.
Given an arbitrary object in R, say the object john, you can ask R whether it is a vector, matrix,
data frame, list or factor by running commands of the form
is.vector(john)
is.matrix(john)
is.data.frame(john)
is.list(john)
is.array(john)
is.factor(john)
Sometimes an object can be coerced into an object of another type. For example, a matrix, say
john, can always be coerced into a data frame like this:
peter <- as.data.frame(john)
The object peter now contains the same data as john, but it is a data frame rather than a matrix.
Similarly one has the coercion operations as.vector, as.matrix, . . . . Be aware that coercion
is not always possible!
8
6
Writing data to plain text files
In R data can be saved in two ways: one can save data in plain text files and one can save data
in a native (binary) format. Both methods have their specific advantages. If you want to save the
content of, say, an object height in a plain text file named foo.txt, then run the following
command
write(height,file="foo.txt")
By this a plain text file foo.txt is created in the working directory of R. This file can be read by
text editors. By default the data in this file is in arranged in 5 columns. If you want to have your
number material arranged in, say, 2 columns, then run
write(height,file="foo.txt",ncolumns=2)
Again the file will be stored in the working directory of R. If you don’t know where this directory
resides, then run
getwd()
to see the path to this folder on your screen.
If you want to have the text file saved on a place outside the working directory, then run something
like
write(height,file="C:/path/to/foo.txt")
The advantage of saving an object in a text file is that these files can be read in not only to R, but
also to other programs, such as SPSS, Matlab, Scilab, Stata, Excel etcetera. These other programs,
like R, are also able to write data to text files. Thus text files present a powerful way to move data
around over different programs. One of the disadvantages of saving data in text files is that one
can save only simple objects in this way and only one per file.
In the next section it is explained how to read in data from text files to R.
7
Reading in data that is stored in plain text files
Suppose your data is stored in a plain text file, named example1.txt. The data could be as
sketched in the figure below:
How to get this in your workspace in R? This can be done by means of the utility read.table().
First of all determine the path to the place where your file resides; this will be something of the
form:
9
C:/path/to/example1.txt
The data can be retrieved from this place by passing the command:
read.table(file="C:/path/to/example1.txt")
Don’t forget the quotations here! You will see the data directly printed to the screen. If you want
to have your data stored in an object named, say mystuff, then pass the following command:
mystuff <- read.table(file="C:/path/to/example1.txt")
The data is now in your workspace and is stored in an object named mystuff.
The method sketched above also works if you want to retrieve data from your USB pendrive. Just
find out which drive letter Windows assigned to your pendrive and then determine the right path
to your file.
Be aware that the above only works for plain text files and not for MS-Word files in document
format!
Next, suppose your data is stored in two columns in a text file, named example2.txt. The data
could be as sketched in the figure below:
How to get this in your workspace in R in a 2-column layout with the headers included? Again
you can use the utility read.table() to bring this about. Suppose that the path to the file
example2.txt is
C:/path/to/example2.txt
Then the data can be retrieved from this place by passing the following command:
mystuff <- read.table(file="C:/path/to/example2.txt",header=TRUE)
The data is now in your workspace and is stored in an object named mystuff. Note that you can
view the first and last six rows of mystuff by running the commands
head(mystuff)
tail(mystuff)
When running into problems during read-in, always check the following:
• Did you type in the right path to your text file?
• Did you use the right slashes in the path to your data file?
• Did you put quotations around the path?
• Did you use the right decimal separators in your dataset?
• If not a space or tab, did you specify the column separator?
10
8
Saving data to native file format
To illustrate how to save data to the R-native format, suppose you have the objects height and
bmi in your workspace. Then you can save them in one file, say, the file foo.RData like this
save(height,bmi,file="foo.RData")
The file foo.RData can now be found in the working directory of R. The file contains the content
of the two objects height and bmi, but the content cannot be read in a text editor. It is a so-called
binary file and its format was specifically designed for usage in R.
In the next section it is explained how data can be loaded from native file format.
9
Loading data that is stored in native file format
Suppose your data is stored in objects that are stored in an R-file named foo.RData. How to get
these objects in R? This is very easy; just pass the command
load(file="X:/path/to/foo.RData")
where X stands for the drive on which the file resides. All the objects saved in the file foo.RData
are then in your workspace. Pass the command ls() to see what you got in your workspace. If
everything went well you can view the first six rows of an object by using the function head().
More precisely, if you want to view the first six rows of an object named, say height, then run
the following command
head(height)
Similarly you can use the function tail() to view the last six rows of an object.
10
Loading built-in data into your workspace
In R there are many of real datasets available. They have been built in during the installation of
the program. To get an impression, just type
data()
Among the list that will appear on the screen there is for example a dataset named ‘faithful’. To
get this dataset in your workspace, type
data(faithful)
Then, for example, type the commands
head(faithful)
tail(faithful)
to view respectively the first and last six rows of the dataset in question.
Of course you would like to know what the dataset is about. To get a description of the meaning
of the columns in the dataset, pass the following command
help(faithful)
11
You will then learn that it is about eruption durations and waiting times as to some geyser in the
Yellow Stone Park in the USA. Note that, when running R under Unix systems, you can quit the
help screen by typing
q
As a simple exercise one could make, for example, a histogram of the waiting times like this
waiting.time <- faithful[,2]
The above selects the second column in the data frame faithful and stores the numbers in the
object waiting.time. A histogram can be made by passing the command
hist(waiting.time)
The histogram below will then appear
30
20
0
10
Frequency
40
50
Histogram of waiting.time
40
50
60
70
80
90
100
waiting.time
11
Importing foreign data
Suppose your data is stored in an SPSS file named foo.sav in the root of drive X. How to get
these data in R? In R there are utilities to bring this about. However, these utilities are not loaded
by default when you start the program. You must first load the library foreign by running the
command:
library(foreign)
Once the library is loaded you have for example the function read.spss() at your disposal.
Via this function the data stored in foo.sav could be loaded, say, into an object john:
john <- read.spss(file="X:/foo.sav",to.data.frame=TRUE)
Similarly there are the functions read.sas(), read.ssd() etcetera.
Type library(help=foreign) to learn about the utilities provided by this library.
12
12
Importing Excel data
Suppose your data is stored in an Excel file named foo.xls in the root of drive C. That is to say,
the path to the file is C:/foo.xls. How to get these data in R? In R there are utilities to bring
this about. However, these utilities are not loaded by default when you start the program. You first
have to load the library RODBC. Before loading this library, however, make sure that it is installed
on your machine. If not, then first install it! Then load the library:
library(RODBC)
Now suppose that you want to store the first three sheets of the Excel file foo.xls in three
objects john, peter and nicholas. This can be done as follows:
chan <- odbcConnectExcel("C:/foo.xls")
john <- sqlFetch(chan,"Sheet1")
peter <- sqlFetch(chan,"Sheet2")
nicholas <- sqlFetch(chan,"Sheet3")
close(chan)
Don’t forget to close the channel! That is to say, don’t forget to finish the operation by the
command
close(chan)
If you forget this, the file foo.xls could get damaged.
The library RODBC provides more utilities then just the import of Excel files. Type
library(help=RODBC)
to get an impression of the tools in this library.
13
Reading in tabulated data
How to read in tabulated data to R? For example, how can one put in data to R that is given in the
form of a frequency table like the one below?
score
A
AB
B
O
frequency
11
8
7
9
One could proceed as follows: First define an object bloodtypes
bloodtypes <- c("A","AB","B","O")
Mind the quotation marks! Then define a frequency vector like this
frq <- c(11,8,7,9)
Then pass the following command
mydata <- rep(bloodtypes,frq)
To see the result of your actions, type
mydata
and strike the key Enter. Next, try to get your frequency table back by passing the command
table(mydata)
13
14
Generating random data
Suppose you want to do some exercises with normally distributed data, but you don’t have such
data at hand. Then R can generate some (artificial) data for you. Just to have something to work
with. As an example, by passing the command
x <- rnorm(100,170,5)
a sequence of 100 numbers is stored into an object x. The sequence will show a normally distributed structure. The mean and the standard deviation of the sequence will be approximately
equal to 170 and 5 respectively.
How is such a sequence generated? This process could be thought to be something like this:
Somewhere in R there is stored a very very long sequence of numbers, a sequence that is perfectly
normally distributed with mean 170 and standard deviation 5. By the command passed above, R
is drawing at random 100 numbers from this sequence. See the section Probability distributions
for more info about the function rnorm() and its relatives.
15
Mean and median
In R it is very easy to obtain the mean or the median of a sequence of numbers. As an example,
type
x <- c(4,2,5,1,4,8,1)
By this the numbers 4, 2, 5, 1, 4, 8, 1 are stored in the object x. To get their mean, pass the command
mean(x)
To get a trimmed mean, type something like
mean(x, trim=0.025)
A total amount of 5% is then trimmed from your data and the mean is taken over the remaining
part. If you want to see the median of the sequence 4, 2, 5, 1, 4, 8, 1, type
median(x)
16
Variance
In R it is very easy to obtain the variance or the standard deviation of a sequence of numbers. As
an example, type
x <- c(4,2,5,1,4,8,1)
By this the numbers 4, 2, 5, 1, 4, 8, 1 are stored in the object x. To get their variance, pass the
command
var(x)
If you want to see the standard deviation of 4, 2, 5, 1, 4, 8, 1, type
sd(x)
Alternatively you could take the square root of the variance by passing the command
sqrt(var(x)).
14
17
Transforming data
Given some sequence of numbers in R, how to transform it? As an example, define a sequence x
by
x <- c(4,2,5,1,4,8,1)
How can one determine a sequence y that contains the squares of the numbers 4, 2, 5, 1, 4, 8, 1?
This can be done as follows:
y <- x^2
Similarly one can determine a sequence z, containing the exponentials of the numbers in x, like
this
z <- exp(x)
In the same way one can transform the sequence x for example by means of the functions log(),
sin(), cos() and so on.
18
Quantiles and boxplots
In R one can quickly obtain obtain quantiles of a sequence of numbers. To illustrate this, first
generate some data and store it into the object x :
x <- rnorm(30,175,10)
By this a sequence of 30 numbers is stored in the object x. To get the quartiles of the sequence,
pass the command
quantile(x)
If you want to see the associated boxplot, type
boxplot(x)
To get the quantile belonging to the fraction 0.35 (or the 35th percentile), type
quantile(x,probs=0.35)
19
Making your own functions
In R virtually all elementary mathematical functions are available by default. Nevertheless you
may very well encounter situations where you need a user-defined function. Suppose, for example,
that you want to have a function
8
x 7→
e−2x
1 + x2
available in R. To make such a function in your workspace, first make a decision how to name the
function. If you decide to name it, say, john then pass the command
john <- function(x) {(8/(1+x^2))*exp(-2*x)}
By this action the function john is available in your workspace. As an example, to get the value
of john for x = 1.2, pass the command
15
john(1.2)
The value of the function john in x = 1.2 is then printed to the screen. The function john
allows for vectorial input. As an example, define a vector myx like this
myx <- c(1.2,1.7,2.2,2.5,2.9)
Now pass the command
john(myx)
This will print a set of function values to the screen. These function values, of course, can also be
stored in, say, an object y :
y <- john(myx)
A plot of the function values could then be obtained like this:
plot(myx,y)
In the section Plotting curves it is explained how you can make nice graphs of arbitrary functions.
20
Saving functions
Functions can be saved for usage in subsequent sessions. Saving it in the root of drive X can be
done like this
save(file="X:/myfun.RData",john)
By this action the function john is saved in a file named myfun.RData. You can save more
than just one function in a file. If there were two more functions in your workspace, say peter
and nicholas, then the three functions could be saved in one file as follows
save(file="X:/myfun.RData",john,peter,nicholas)
By this the two functions are saved in a file named myfun.RData in the root of drive X. In a
subsequent session you can load the two functions in your workspace like this
load("X:/myfun.RData")
The functions john,peterand nicholas are then again at your disposal.
The code (or syntax) that defines functions can also be stored in plain text files. For example, by
means of a text editor one could save the code
john <- function(x) {(8/(1+x^2))*exp(-2*x)}
peter <- function(x) {0.5+log(x/(1+x))}
nicholas <- function(x) {1+sin(3*x)}
in a text file named myfun.txt. On a Windows machine this can be done, for example, in the
program ‘WordPad’. It can also be done in the program ‘MS-Word’ but then you must save the
result as a plain text file, not as a file in MS document format! To read in your function from the
text file myfun.txt to your workspace, pass the command
source(file="X:/path/to/myfun.txt")
Here X is the drive on which the file myfun.txt resides, followed by the complete path to this
file.
16
21
Plotting curves
Suppose you have produced some graphics with R and then you decide to add some curve to it. In
R this can be done by means of the plot utility curve(). As a first example, suppose that you
want to plot the graph of the function x 7→ exp(−x) over the interval [0, 6]. This can be done as
follows:
curve(exp(-x),xlim=c(0,6))
Next, you can add the graph of the sinus function to the plot created above like this
curve(sin(x),xlim=c(0,6),add=TRUE)
Normal curves can be plotted by means of the function dnorm(). For example
curve(dnorm(x,160,10),xlim=c(130,180))
plots a normal curve with mean µ = 160 and standard deviation σ = 10. The plot utility
curve() can also be used to add normal curves to a histogram. To illustrate this, first generate some data to work with:
height <- rnorm(100,175,5)
Next, make a histogram of the object height by running the command
hist(height,freq=FALSE)
The option freq=FALSE is used here to indicate that, rather than setting out the frequencies on
the vertical axis, this axis must be scaled such as to result in a histogram with total area equal to
1. A normal curve (of maximal fit to the histogram) can be added as follows. First store the mean
and the standard deviation of height in two objects, for example in m and s, like this
m <- mean(height)
s <- sd(height)
Then pass the command
curve(dnorm(x,m,s),add=TRUE)
Of course you can also plot the functions you created yourself. If you created for example the
functions john, peter and nicholas, as described in the section Making functions, then you
could plot them as curves in one figure like this
curve(john(x),xlim=c(1,5))
curve(peter(x),xlim=c(1,5),add=TRUE)
curve(nicholas(x),xlim=c(1,5),add=TRUE)
As another example, a normal density curve could also be plotted by first making a user-defined
function nicholas :
nicholas <- function(x) {dnorm(x,160,10)}
To plot it, pass the command
curve(nicholas(x),xlim=c(130,180))
Just try it out to feel how it works!
17
22
Plotting straight lines
Suppose you have produced some graphics with R and then there is suddenly a wish to add some
straight line to your graph. This can be done via the function abline(). As an example to show
how it works, first pass the commands
u <- c(1,2,3,4)
v <- u^2
plot(u,v)
This will create a scatter plot with 4 points. Adding to this plot the straight line with equation
y = 1 + 2x
can be done like this
abline(1,2)
After striking the key Enter the line will be there! To plot a vertical line, at level x = 2.5, do the
following:
abline(v=2.5)
Similarly, a horizontal line at level y = 12 can be added like this
abline(h=12)
Note that the above could also be brought about by passing the command abline(12,0).
23
Kolmogorov-Smirnov plots
How to make a KS-plot in R? To illustrate this, consider the flood size data of the Ocmulgee River
and store it into the (vectorial) object flood like this
flood <- c(50,12,16,20,17,13,61,26,33,38)
To make the empirical cumulative frequency function (ecdf) that belongs to the data, run the
command
john <- ecdf(flood)
To get a plot, do this
plot(john,verticals=TRUE)
If you want to include labels on the x-axis, then do something like
plot(john,verticals=TRUE,xlab="Size of Flood")
Similarly a label can be attached to the y-axis.
To add the associated normal cdf to the plot created above, procede as follows: First store the
mean and the standard deviation of flood in two objects, for example in m and s, like this
m <- mean(flood)
s <- sd(flood)
Then pass the command
18
curve(pnorm(x,m,s),add=TRUE)
0.8
0.6
0.4
0.2
Cumulative Proportion
1.0
The figure below will then be created
20
30
40
50
60
Size of Flood
24
Kolmogorov-Smirnov distance
The KS-distance is a component in the so-called Kolmogorov-Smirnov test, which will be discussed in more detail in a later section. To show how the KS-distance to normality can be obtained, store (to have some data) the flood size data of the Ocmulgee River into the object flood
as follows:
flood <- c(50,12,16,20,17,13,61,26,33,38)
Then store the mean and the standard deviation of flood in two objects, for example in m and s,
like this
m <- mean(flood)
s <- sd(flood)
Finally, run the command
ks.test(flood,"pnorm",m,s)
The following output will then be returned
19
The KS-distance is denoted by the character D and can be read off to have the value 0.1968. In the
output you can also find the p-value.
25
QQ-plots
QQ-plots can easily be made in R. As an example, generate some data to work with:
x <- rnorm(50,175,10)
Then pass the command
qqnorm(x)
The result will be a graph like the one below:
180
170
160
Sample Quantiles
190
Normal Q−Q Plot
−2
−1
0
1
2
Theoretical Quantiles
A reference line in your QQ-plot can be added by passing the command (presumed your variable
is named x)
abline(mean(x),sd(x))
The QQ-plot will then look like this
20
180
170
160
Sample Quantiles
190
Normal Q−Q Plot
−2
−1
0
1
2
Theoretical Quantiles
26
Probability distributions
In R you can deal with (for example) the t-distributions via the functions
dt()
pt()
qt()
rt()
The function dt() returns values of t-densities. You could use this function, for example, to make
a graph of a t-density:
curve(dt(x,8),xlim=c(-3,3))
This would create a graph of the t-density with 8 degrees of freedom.
The function pt() returns the cumulative probabilities that emanate from t-distributions. For
example, the command
pt(1.32,9)
would return the size of a left probability tail with cut-off point 1.32 in a t-distribution with 9
degrees of freedom.
The function qt() does the converse of the function pt(). So the command
qt(0.95,12)
returns the 95th percentile of a t-distribution with 12 degrees of freedom.
Finally, the function rt() generates samples from t-distributed populations. For example, the
command
21
rt(100,12)
would return a sample of length 100 from a t-distributed population with 12 degrees of freedom.
A scheme of probability distributions and their abbreviations in R is given below:
Distribution
Normal
Student’s T
Binomial
Poisson
Gamma
Beta
Uniform
Exponential
Geometrical
Fisher
χ2
Abbreviation
norm
t
binom
pois
gamma
beta
unif
exp
geom
f
chisq
These distributions can all be used in combination with the prefixes d, p, q and r.
Prefix
d
p
q
r
Function
returns values of the density
returns cumulative probabilities
returns quantiles
returns random samples
For example, for normal distributions one has the functions dnorm(), pnorm(), qnorm() and
rnorm().
27
Plotting probability densities
Suppose you want a plot of the χ2 - density with 3 degrees of freedom over an x-interval that runs
from 0 to 18. Then just pass the command
curve(dchisq(x,3),xlim=c(0,18))
If you want to add the χ2 - density with 5 degrees of freedom to the plot, then run
curve(dchisq(x,5),add=TRUE)
Perhaps you want have the graph of the χ2 - density with 10 degrees of freedom also added to the
plot. Then just run
curve(dchisq(x,10),add=TRUE)
In this way χ2 - densities can also be added to histogram plots.
If you want to plot the cumulative distribution function of a χ2 - distribution, then do the same as
sketched above, thereby replacing the function dchisq() by pchisq().
22
28
Confidence intervals for a population mean
Confidence intervals can be obtained by applying a t-test. To illustrate one thing and another, make
some artificial data. For example, some artificial weight data
weight <- c(3.7, 7.4, 8.6, 5.3, 4.4)
To carry out a t-test with null hypothesis
H0 : µ = 6.8
pass the command
t.test(weight, mu=6.8)
Then strike the key Enter. The output will then immediately appear on the screen. It is of the
form depicted below:
In the output above you can see that a 1-sample t-test was carried out (t-tests come in several
flavours). As you can see, the output also contains a 95% confidence interval for the population
mean. It is the interval (3.321148; 8.438852). If you are only interested in a confidence interval,
then it is not important which null hypothesis you choose.
29
The 1-sample t-test
In R all flavors of the t-test can be conducted by means of the command t.test(). To illustrate
how a 1-sample t-test can be carried out, first make some artificial height data to work with
height <- c(174,165,179,189,172,161)
To test for example the null hypothesis
H0 : µ = 175
just pass the command
t.test(height,mu=175)
In the output that is returned (after striking the key Enter) you can find for example a p-value
and a confidence interval for the population mean.
23
30
The 2-sample t-test
To illustrate how to conduct a 2-sample t-test you need two independent datasets. Make them in
an artificial way and think of them as being height measurements in Holland and France:
holland <- c(174,165,179,189,172,161)
france <- c(173,180,172,155,169)
To figure out whether these two samples differ significantly in mean, just pass the command
t.test(holland,france)
Among the output there is a p-value and a confidence interval for the difference in population
means in Holland and France.
If the heights would have been put in as one vector, say, a vector height with a grouping variable
group to define the groups, then the t-test can simply be run as
t.test(height~group)
31
The paired-sample t-test
A paired-sample t-test can be conducted in R in very much the same way as a 2-sample t-test. To
illustrate this, first make some data to work with
height1 <- c(174,165,179,189,172,161)
height2 <- c(175,167,179,189,173,162)
Think of these data as being six pairs of height measurements. For example, the height of six
adolescents could be measured at two points in time. To check whether there is a significant
change in height, pass the following command
t.test(height1,height2,paired=TRUE)
In the output that is returned (after striking the key Enter) you can find a p-value and a confidence
interval for the mean difference in height.
32
Kolmogorov-Smirnov tests
To show how a KS-test on normality can be run in R, store (to have some data) the flood size data
of the Ocmulgee River into the variable flood as follows:
flood <- c(50,12,16,20,17,13,61,26,33,38)
Then store the mean and the standard deviation of flood in two variables, for example in m and
s, like this
m <- mean(flood)
s <- sd(flood)
Finally, run the command
ks.test(flood,"pnorm",m,s)
R will return some info then, including a p-value and the KS-distance.
24
The KS-distance is denoted by the character D and can be read off to have the value 0.1968. The
p-value is 0.7653.
If, for certain reasons, you would like to check whether the sample flood could have had its
origin in a t-distributed population with 6 degrees of freedom, then pass the command
ks.test(flood,"pt",6)
R will return a KS-distance and a p-value.
Similarly, if you want to check whether flood is a sample from an exponentially distributed
population with mean 3, run the command
ks.test(flood,"pexp",3)
R will return a KS-distance (to an exponential distribution with parameter 3) and a p-value.
8 -distribution,
If you want to check whether the distribution of flood differs significantly from an F10
pass the following command:
ks.test(flood,"pf",8,10)
8 -distribution) and a p-value.
R will return a KS-distance (to an F10
33
Fisher’s variance test
Suppose you have in your workspace two sequences of measurements as to lung capacity of 8-year
old children. One sequence, say, lungcapgirls contains the measurements of the girls and the
other, lungcapboys, those of the boys. After inspection it appears that both sequences pass the
KS-test as to normality. Next you want run Fisher’s variance test to check whether the variance in
both sequences differs significantly. In R you can do this by simply passing the command
var.test(lungcapgirls,lungcapboys)
After striking the key Enter you will have a p-value returned.
If all lung capacities of the girls and boys were stored in one column, say a column lungcap,
and if the groups were defined by a grouping variable, say a variable named sex, then one could
run Fisher’s variance test as follows:
var.test(lungcap~sex)
Th test results will be returned after striking the key Enter.
25
34
Splitting datasets
Suppose the dataset d below is given:
You want to split up the dataset into two datasets d1 and d2 according to the grouping variable
sex. This can be done as follows by using the function subset() :
d1 <- subset(d,sex==1)
d2 <- subset(d,sex==2)
To extract those rows for which the variable bmi is larger than 22, run
d3 <- subset(d,bmi>22)
Another way to split the dataset d is by using the function split() in the following way:
attach(d)
ds <- split(d,sex)
By this action a so-called ‘list’ is created (see the section Numbers, strings and objects in R).
The list contains two components: a component ds$1 and a component ds$2. To extract the
measurements with sex codes 1 and 2 as separate datasets from the list, just pass the command
d1=ds$1
d2=ds$2
Note that if the coding were "male" and "female" rather than 1 and 2, then the components
of the list ds would have been ds$male and ds$female.
35
The 1-way ANOVA
Running ANOVA’s in R is very easy. To give an example, load some built-in data into you
workspace:
data(PlantGrowth)
This dataset arose from an experiment on plant growth under several conditions. There is a wish
to check whether the conditions resort significantly in effect. To bring this about, first have a look
at the headers of the dataset (they are weight and group). If you want to use these headers as
if they were variables in your workspace, pass the command:
26
attach(PlantGrowth)
Then run the following command (where PG stands for Plant Growth):
PG <- lm(weight~factor(group))
By this action R ran the ANOVA mentioned above. The results are stored in an object which you
gave the name PG. To view the results, type
anova(PG)
A list of residuals can be extracted from the object PG like this
e <- residuals(PG)
By the above the residuals of the ANOVA are stored into the variable e. This variable e can then
be checked on normality.
To check homogeneity of variances one could run Bartlett’s test:
bartlett.test(weight~factor(group))
The null hypothesis in Bartlett’s test is that there is homogeneity indeed. Bartlett’s test is a generalization of Fisher’s variance test for situations in which there are more than two groups. Both
Bartlett’s and Fisher’s test lean on normality assumptions and are very sensitive to deviations in
this. A more robust alternative to these tests is Levene’s variance test. This test is part of the
library car. Once this library is loaded, Levene’s test could, in this example, be run by passing
the command
levene.test(PG)
Equivalently one could run
levene.test(weight~factor(group))
Be aware that the p-values returned by Bartlett’s and Levene’s test may be very different! Also be
aware that R uses a version of the Levene’s test that differs from the version used in, for example,
SPSS.
In R multiple comparisons can be done like this
pairwise.t.test(weight,factor(group))
This action will return a kind of a cross table from which the corrected p-values of all comparisons
can be read off.
36
The 2-way ANOVA
A 2-way ANOVA in R can be run similar to a 1-way ANOVA. To give an example, load some
built-in data into you workspace:
data(morley)
27
This is the dataset that emerged from the famous Michelson-Morley experiments, in which the
speed of light was measured. There is a wish to check whether the factors Run and Expt resort
effect. Interaction is excluded beforehand. To bring this about, first have a look at the headers
of the dataset (they are Expt, Run and Speed). These headers can be revealed by using the
function head() or tail(). If you want to use these headers as if they were variables in your
workspace, pass the command:
attach(morley)
Then run the following command (where MM stands for Michelson-Morley):
MM <- lm(Speed~factor(Expt)+factor(Run))
If you want to include the possibility of interaction in your 2-way ANOVA, then run the command
MM <- lm(Speed~factor(Expt)+factor(Run)+factor(Expt):factor(Run))
Equivalently, the above could be run by passing the command
MM <- lm(Speed~factor(Expt)*factor(Run))
However, on this dataset this would not work: For a 2-way ANOVA with interaction each cell
must contain at least two elements. This condition is not met here ...
Levene’s test (in the library car), to check homogeneity of variance, could be carried out by
simply passing the command
levene.test(MM)
Equivalently one could run
levene.test(Speed~factor(Expt)*factor(Run))
Again, for this particular dataset there is a problem in running the above because each cell contains
only one measurement. For this reason one cannot talk about the variance of the measurements in
a cell.
37
Wilcoxon’s signed-rank test
First make some suitable data to work with
x <- c(0.80, 0.83, 1.89, 1.04, 1.45)
y <- c(1.15, 0.88, 0.90, 0.74, 1.21)
Assume that the data in these two sequences are paired, that is to say, 0.80 belongs to 1.15, 0.83
belongs to 0.88 and so on. Run a Wilcoxon test by passing the command
wilcox.test(x,y,paired=TRUE)
If you want exact p-values, run
wilcox.test(x,y,paired=TRUE,exact=TRUE)
28
38
The Mann-Whitney test
In R the Mann-Whitney test is referred to as a Wilcoxon rank-sum test. To illustrate how to run
this test, first make some suitable data to work with:
x <- c(0.80, 0.83, 1.89, 1.04, 1.45, 1.38)
y <- c(1.15, 0.88, 0.90, 0.74, 1.21)
Then run the command
wilcox.test(x,y)
To give another example, load the built-in dataset sleep into your workspace
data(sleep)
This dataset contains two columns with headers extra and group. The column with header
group serves as a grouping variable indeed. If you want to use the headers as if they were
variables in your workspace, pass the command
attach(sleep)
Then run a Mann-Whitney test like this
wilcox.test(extra~group)
If you want exact p-values, run
wilcox.test(extra~group,exact=TRUE)
39
The Kruskal-Wallis test
First load some suitable built-in dataset:
data(chickwts)
This dataset contains two columns with headers weight and feed. The column with header
feed could serve here as a grouping variable. If you want to use these headers as if they were
variables in your workspace, pass the following command
attach(chickwts)
To run a Kruskal-Wallis test, pass the command
kruskal.test(weight~feed)
The Kruskal-Wallis test can be regarded as a nonparametric alternative to a 1-way ANOVA.
40
The runs test
The runs test is not contained in R-base. You will have to load the library tseries. Once having
done this, it is possible to do a runs test in R. Given a sequence of symbols x one can do it like
this:
runs.test(x)
If R starts complaining that the input is not a factor, then pass the command:
runs.test(factor(x))
R will return some info, among which a p-value.
29
41
Simple linear regression
To learn how to run a simple linear regression in R, put in the following dataset:
x <- c(1,2,3,4,5)
y <- c(3,3,5,7,7)
Then pass the command
lm(y~x)
The regression coefficients will be returened by this action. Wiser than the above would be to run
a command of type
myregression <- lm(y~x)
By this action all results of the regression analysis (the residuals included) are stored into an object
that you have given the name myregression. A summary of the whole regression analysis can
be obtained like this
summary(myregression)
Confidence intervals for the coefficients can be extracted from the object myregression by
passing the command
confint(myregression)
To extract from the object myregression the residuals, pass the command
e <- residuals(myregression)
By the above the residuals are stored into an object e and can then be used for further analysis.
To get the fitted values stored in an object yhat do the following
yhat <- fitted(myregression)
To get a prediction for the value of y when it is given that x = 3.5, first store the value of 3.5 in
the object x0 like this:
x0 <- data.frame(x=3.5)
Then run the command
predict(myregression,newdata=x0,interval="pred")
This will provide you with the desired prediction intervals.
To illustrate how you can run a regression through the origin, type in some data to work with
x <- c(1,2,3,4,5)
y <- c(3,3,5,7,7)
Recall that an ordinary simple regression can be run by passing the command
myregression <- lm(y~x)
All results of the regression analysis are then stored into an object that you have named myregression.
To run a simple linear regression through the origin, pass the command
myregression <- lm(y~x-1)
All information can then be extracted from the object myregression as was sketched above.
30
42
Multiple linear regression
In the previous section it was explained how to run a simple linear regression in R. If you want to
explain your variable y by two x-variables in your regression model, say,
x1 and x2, then pass the command
myregression <- lm(y~x1+x2)
If you don’t want the constant in your model, this can be brought about as follows:
myregression <- lm(y~x1+x2-1)
You are then running a multiple linear regression through the origin. If an x-variable, say x2,
happens to categorical, then run
myregression <- lm(y~x1+factor(x2))
43
Regression plots
In the following it is assumed that a simple regression analysis was already run:
myregression <- lm(y~x)
A basic regression plot consists of a scatter plot with the regression line in it. This can be obtained
like this:
plot(x,y)
abline(myregression)
A residual plot can be obtained as follows:
e <- residuals(myregression)
yhat <- fitted(myregression)
plot(yhat,e)
plot(myregression)
The function segments() plots adds line segments to scatter plots. This function can also be of
use in regression plots. To discover how it works, define
u1
u2
v1
v2
<<<<-
1
4
5
1
Then pass the command
plot(c(u1,u2),c(v1,v2))
segments(u1,v1,u2,v2)
You will see that the function segments() adds line segments to a scatter plot.
A scatter plot in which the regression line and the residuals are visible can be obtained like this
plot(x,y)
segments(x,y,x,yhat)
Type ?segments to get some more info about the function segments().
31
44
Generating random regression data
The three conditions in linear regression are:
• In the population y must be quasi linear in x.
• In the population the residuals must be normally distributed.
• There must be homoscedasticity.
How to generate random regression datasets that satisfy the above conditions? For example, how
can one generate samples from a population with regression line y = 2 + 3 x? This can be done
as follows. First create a sequence of x -values, for example like this:
x <- 1:25
Then create the corresponding y -values:
y <- 2 + 3*x + rnorm(25,0,2)
Now the variables x and y define 25 points in the plane. These 25 points can be thought to
be a sample from a population that meets the three regression conditions. The population has a
regression line y = 2 + 3 x and the residuals relative to this line are normally distributed with
standard deviation 2.
Similarly one can create such datasets in multiple linear regression. As an example, one could
create
x1 <- 1:24
x2 <- rep(1:12,2)
y <- 2 + 3*x1 - x2 + rnorm(24,0,3)
The variables x1, x2, y define 24 point in 3-dimensional space. These 24 points can be thought
to be a sample from a population that meets the three regression conditions. The population has
a regression equation y = 2 + 3 x1 − x2 and the residuals relative to this equation are normally
distributed with standard deviation 3.
45
Binomial tests
A tool to run a 1-sample proportion test in R is the binomial test. To illustrate how to use it, suppose
that in an experiment you have drawn a sample of 86 patients from a certain population. Among
these 86 patients there are 69 smokers. You want to test the null hypothesis that the population
proportion of smokers be 70%. This can be done like this
binom.test(69,86,0.70)
The output below will then be returned
From the output it can be read off that the the p-value is 0.04468.
32
46
Proportion tests
A versatile and userfriendly tool to analyze proportions in R is the proportion test. To illustrate
how to run a 1-sample proportion test, suppose that in an experiment you have drawn a sample of
86 patients from a certain population. Among these 86 patients there are 69 smokers. You want
to test the null hypothesis that the population proportion of smokers be 70%. This can be done as
follows:
prop.test(69,86,0.70)
Note that the above could also be settled with the binomial test. Actually, the advantage of the
binomial test is that it returns exact p-values. The proportion test returns approximate p-values.
The advantage of the proportion test, however, is that it is more versatile. It can, for example, be
used to compare two (or even more than two) proportions. To show how the proportion test can
be used to compare two empirical proportions, suppose that you have two populations of patients,
say population A and population B. From population A you draw a sample of 86 patients and a
count reveals that there are 69 smokers among them. Similarly, a sample of 93 patients is drawn
from population B and it turns out that there are 85 smokers among them. A test to check whether
the two proportions 69:86 and 85:93 differ significantly can be run like this
smokers <- c(69,85)
patients <- c(86,93)
prop.test(smokers, patients)
To compare three smoker proportions, for example 69:86, 85:93 and 121:136, do the following
smokers <- c(69,85,121)
patients <- c(86,93,136)
prop.test(smokers, patients)
47
Contingency tables
How to make contingency tables (matrices) in R? Just try out the following
x <- c(794,86,150,570)
y <- matrix(x,ncol=2)
The rows in matrix y can be assigned names in the following way:
rownames(y) <- c("first row","second row")
Similarly the columns can be assigned names
colnames(y) <- c("first column","second column")
Print the variable y to the screen to view the effect of the actions described above.
48
McNemar’s test
The McNemar test is easy to run in R. To have some data, put in a table like this:
x <- c(12,14,4,20)
y <- matrix(x,ncol=2)
33
Edit the contingency table for example like this, where ‘yes’ stands for ‘yes, the person sees relief’
and ‘with’ for ‘with 3D-glasses’:
colnames(y) <- c("yes-with","no-with")
rownames(y) <- c("yes-without","no-without")
Your contingency table is now ready for use. When you print the matrix y to the screen you will
see something like this:
McNemar’s test can now be run in this way:
mcnemar.test(y)
On the screen there will then be something returned like this
49
The chi-square test on goodness of fit
To learn how to run a χ2 - test on goodness of fit, use the data below as an example:
Observed
Expected
A
56
68
B
29
30
AB
58
46
O
57
56
Store the observed frequencies in a variable which you give for example the name observed:
observed <- c(56,29,58,57)
Then convert the expected frequencies into a so-called probability vector named (for example)
expected, like this
expected <- c(68,30,46,56)
expected <- expected/sum(expected)
Print the vector expected to the screen to see the result. To run the χ2 - test, pass the command
chisq.test(observed,p=expected)
34
50
The chi-square test on independence
How to run a χ2 - test in R on the next contingency table?
European Shorthair
Russian Blue
A
107
43
B
36
24
AB
17
13
First you have to put in the table. This can be done, for example, like this
ES <- c(107,36,17)
RB <- c(43,24,13)
x <- rbind(ES,RB)
When printing x to the screen you will see something like this
If you want (this is optional) you can assign headers to the columns, for example like this
colnames(x) <- c("A","B","AB")
Print x to the screen to view the result of this action. To run a χ2 - test on the newborn contingency
table x, simply pass the command
chisq.test(x)
The output will be something like this
35
The p-value can be read off to be 0.134.
The null hypothesis is therefore maintained, that is to say, the dataset provides no evidence that
there is a relation between blood type and the two cat races.
In a very similar way one could run Fisher’s exact test on the contingency table x:
fisher.test(x)
The output below will then be returned
Note that the p-values returned by Fisher’s exact test are exact indeed, whereas those returned by
the χ2 - test only approximate.
51
Fitting nonlinear models
To illustrate things one could start from the dataset below:
x
1
2
3
4
5
y
2
3
5
5
9
Start the program R and put in the above dataset (from console) as follows:
x <- c(1,2,3,4,5)
y <- c(2,3,5,5,9)
Then strike the key Enter. Next, pass the command
nls(y~A*(x**B)+C,start=list(A=1,B=2,C=-6))
After striking the key Enter you will see the output appear like this:
36
The values for A, B and C can be read off to be 0.1564, 2.3193, 2.1265 respectively and the
minimum value for SSE (presented here as the residual sum-of-squares) is evaluated
as 2.0118.
More info about the routine nls() in R can be obtained by just typing
?nls
and then strike the key Enter. A help screen will then appear. Under Unix systems, type q to
leave this help screen and continue in R.
52
Residual analysis
To carry out a residual analysis in R, first get your variables (say x and y) in the work space. There
are many ways to do this. Just choose one. Now suppose that you want to fit the model
y = A ∗ e−B∗x
Then choose a set of initial values, for example A = 1, B = 2 and pass the command
r <- nls(y~A*exp(-B*x),start=list(A=1,B=2))
After striking the key Enter, an object r is created in the work space of R. This
object contains all results of the model fit, for example also the residuals. To extract the residuals from the object r, pass the command
res <- residuals(r)
Now the residuals are stored in the variable res. To see the content of the variable res on the
screen, simply type
res
and strike the key Enter. There are your residuals!
Checking the residuals on normality can be done via the Shapiro-Wilk test like this
shapiro.test(res)
Homoscedasticity can be checked graphically by creating a residual scatter plot:
plot(x,res)
37
53
Logistic regression
Consider the dataset
height
171
161
183
168
172
189
162
179
problem
1
0
1
0
1
0
0
1
Put the dataset in like this
height <- c(171,161,183,168,172,189,162)
problem <- c(1,0,1,0,1,0,0)
In R a logistic regression can be run (in this particular case) by passing the command
name <- glm(problem~height,family=binomial)
A summary of the results of the analysis can be obtained like this
summary(name)
If, besides the variable height, the dataset also contained a variable BMI, then one could include
this variable in the model like this:
name <- glm(problem~height+BMI,family=binomial)
If there were also a categorical variable blood in the dataset, put it into the model like this
name <- glm(problem~height+BMI+factor(blood),family=binomial)
Model reduction can be applied by running
step(name)
To get a prediction for a person with height 165 cm, a BMI of 24 and bloodtype O, create for
example a variable preddata like this
preddata <- data.frame(height=165,BMI=24,bloodtype="O")
Then pass the command
predict(name,newdata=preddata,type="response")
If in the command above newdata is not specified, then R will return the probabilities relative to
the cases in the dataset to which the fit was carried out.
If you want to extract the value of the likelihood of the fitted model from the object name, then
run the command
logLik(name)
The natural logarithm of the likelihood value will then be returned.
38
54
Nonparametric survival analysis
Consider the following survival data
years
71
61
83
68
72
89
62
79
event
1
2
1
3
1
2
3
1
The events are classified in three categories. If the event of interest is coded by 3 then a corresponding survival object, say mary, can be created like this
mary <- survfit(Surv(years,event==3))
The events coded by 1 or 2 are now regarded as being censored. A Kaplan-Meier plot can be
extracted from the object mary as follows:
plot(mary)
If you don’t want the confidence bands in the plot, then remake the object mary, thereby using
the option conf.type = "none".
Next, suppose that there is a grouping variable in the dataset:
years
71
61
83
68
72
89
62
79
event
1
2
1
3
1
2
3
1
group
1
1
1
1
1
2
2
2
If you want to compare the Kaplan-Meier curves of the groups in a logrank test, then make an
object, say jane, as follows
jane <- survdiff(Surv(years,event)~group)
If you don’t specify the event of interest, then by default the event coded by 1 is taken. By simply
typing jane you will see the results of the logrank test on your screen.
55
Power and sample size (I)
In R there there are some extremely useful functions as to determining power, sample sizes
etcetera. One of them is designed for t-tests. It is the function power.t.test(). To illustrate how to use this function, suppose you want to test a (null) hypothesized population mean
µ = 120. Your sample size is 20 and you want to test on a significance level of α = 0.05. It is
known that the standard deviation of the population is 10. You want to know the power of the test
against the alternative µ = 123. To get this info, run
39
power.t.test(n=20,delta=3,sd=10,sig.level=0.05,type="one.sample")
In the arguments of power.t.test(), n is the sample size, delta is the difference between
the null hypothesized and the alternative mean, sd is the standard deviation of the population and
type is the type of the t-test. You will then get the next output:
From this output you can read off that the power in the alternative µ = 123 is 0.2465.
The function power.t.test() can also be used to compute sample sizes necessary to get
things significant. Suppose, for example, that the null hypothesized population mean is µ = 120.
You expect in your sample a mean of about 125, a standard deviation of about 7 and you want
to have things significant on the 5% level and you want a power of 0.8 or more in the alternative
µ = 125. What sample size must be chosen to have the sample mean significantly different from
the null hypothesized mean? Run the next command to find it out:
power.t.test(delta=5,sd=7,sig.level=0.05,power=0.8,
type="one.sample")
In the output that is returned after striking the key Enter you can find a minimum value for the
sample size in question.
Similar to the function power.t.test() there are the functions power.anova.test(),
power.prop.test(). They can be used when dealing with power and sample size problems
in ANOVA and proportion tests.
56
Power and sample size (II)
In the previous section it was described how to use the three functions power.t.test(),
power.prop.test() and power.anova.test(). More extensive tools to compute power
and sample sizes are provided by the library pwr. This library is not contained in R-base; it must
be installed before using it. To install it, first make sure that you are connected to the internet.
Then run R with Administrator privileges and pass the command
install.packages("pwr")
After a successful installation the library can be loaded into R by running the command
library(pwr)
The contents of the library can be viewed by passing the command
40
help(pwr)
Among the functions in the library pwr there is the function pwr.t.test(). This function can
be used in very much the same way as the function power.t.test(). As an example, one
could run
pwr.t.test(n=20,d=3,sig.level=0.05,type="one.sample")
In the arguments of pwr.t.test(), n is the sample size, d is the effect size (the difference
between the null hypothesized and the alternative mean divided by the standard deviation) and
type is the type of the t-test.
Another tool in this library is the function pwr.2p2n.test(). Use this function when dealing
with 2-sample proportion tests with possible unequal sample sizes. A call to this function could
be as follows:
pwr.2p2n.test(h=0.2,n1=300,sig.level=0.05,power=0.8,
alternative="two.sided")
Here h stands for the difference between the two proportions that you want to have
significant, n1 stands for the size of one of the samples and the other parameters were already
discussed before. The output to this command is depicted in the figure below:
It can be read off that the second sample must be of a length n2 of at least 568 to meet the
requirements specified by the remaining parameters.
41
Appendices
A
Installation of R-base and R-libraries
Installation of R-base
Windows Under Windows operational systems R can easily be installed by downloading a Windows binary installer from the CRAN website. The installer is an executable file named
R-xxx-win32.exe, where xxx stands for the version. Once downloaded, double-click the file
to start the installation. The installation is a straight forward procedure.
MacIntosh Under MacIntosh operational systems one must distinguish between MacOS and MacOSX. Binary installers can be downloaded from the CRAN website.
Linux For a lot of Linux systems there are binary installers, which can be retrieved from the
CRAN website. On some Linux systems R participates in the native package managements.
If so, R can usually be installed via a few mouse clicks. If you have the necessary compilers
installed, then you can also consider to install R to your machine from source code. The
source code can be downloaded from the CRAN website. It is a so-called tarball, named
R-xxx.tar.gz, where xxx stands for the version number. You must carry out the installation
as root!
FreeBSD Under a FreeBSD operational system it is attractive to install R from source code via
the (formidable) ports system. First make sure you have the ports system installed. If so,
then start a console and change directory to the folder /usr/ports/math/R like this:
cd /usr/ports/math/R
Once there, type (as root) the command
make install clean
That’s all. Installation from source code may take a lot of time (depending on your system it
could take several hours). For the impatient there is also a possibility to install a precompiled
binary distribution like this
pkg_add -r R
Make sure you have root privileges and an active internet connection!
The above also applies, grosso modo, to a lot of other BSD flavors.
Installation of R-libraries
R-libraries are extensions to R-base. They can be loaded into R, provided they are installed on
your system. An easy and platform independent way to install R-libraries is as follows:
Step 1 Make sure you have administrator (or root) privileges.
Step 2 Start R as an administrator (or root).
42
Step 3 Under R type the command
install.packages("nameoflib")
where nameoflib stands for the name of the library, for example car or pwr. Don’t
forget the quotation marks! If the library nameoflib depends on other libraries, then it
may be comfortable to run, rather than the above, the command:
install.packages("nameoflib",dependencies="Depends")
A panel will pop up and you will be asked to choose a server from which you want to
download the library.
Step 4 Choose a server and click the button OK. The installation will start then.
Once the library has been installed, it can be loaded by passing the command
library(nameoflib)
The library and all the tools it provides are then at your disposal. Note that loading a library is
quite different from installing a library!
On Windows systems the above procedure can partly be carried out by mouse clicking.
Another way to install a library is to first download the library from
• Libraries for Windows (http://cran.r-project.org/bin/windows/contrib/r-release/)
• Libraries for MacOSX (http://cran.r-project.org/bin/macosx/universal/contrib/r-release/)
• Source code of libraries for Unix (http://cran.r-project.org/src/contrib/)
Under a Windows operational system the library can then be installed by starting R as an administrator and type
install.packages("X:/path/to/nameoflib_xxx.zip",repos=NULL)
The option repos=NULL tells R that it should, rather than searching the web for repositories that
contain the file nameoflib_xxx.zip, just use the file that resides on the place indicated by
your path.
Under a MacIntosh operational system the library can be installed by starting R as an administrator
and then type
install.packages("X:/path/to/nameoflib_xxx.tgz",repos=NULL)
Under Unix systems the library can be installed by starting R as root and then type
install.packages("X:/path/to/nameoflib_xxx.tar.gz",repos=NULL)
Under Unix systems the library can also be installed by running (as root) the following command
in a console:
R CMD INSTALL /path/to/nameoflib_xxx.tar.gz
By this the library will be compiled to your machine from source code. You must have the necessary compilers installed on your system to bring this about!
Libraries can be deinstalled from your system by using the utility remove.packages().
43
B
B.1
How to make your own libraries under R?
Building an R-library under Windows
How to build a library for R under Windows? This section will hopefully lead you through the
obstacles that people usually encounter.
B.1.1
Installation of compilers
In order to build (binary) libraries that can be installed under R under Windows, you need to have
the necessary software installed. To this end you may down load the latest version of Rtools
from http://cran.r-project.org/bin/windows/Rtools/. Once Rtools is installed you have everything on board to create your own libraries under R .
B.1.2
Compiling a binary installer for an R-library under Windows
Step 1 Load the stuff for your library into the R-workspace.
Step 2 Pass (under R) the command
package.skeleton(name="johnvanderbrook", list=ls())
By this action a folder, named johnvanderbrook, is created in the working directory of
R. This folder is called the skeleton of the library that is to be built.
Step 3 The skeleton contains the files DESCRIPTION, read-and-delete-me and the subfolders data, man, R and src. The file DESCRIPTION must be edited by means of a text
editor (for example WordPad, not in MS-Word). The editing could be something like this:
Package: broekjan
Type: Package
Title: What the package does (short line)
Version: 0.9
Date: 2008-05-12
Author: Wiebe Pestman
Maintainer: Who to complain to <[email protected]>
Description: More about what it does (more lines allowed)
License: GNU
The files in the subfolders data and man must also be edited. They contain files with
names of type blabla.Rd. These files are meant to build the help functions and they must
be edited as if they were simplified LATEX files. Knowledge of LATEX is not required in this.
Step 4 If your skeleton is ready, localize the folder where the executable Rcmd.exe resides.
Move or copy your skeleton to this directory.
Step 5 Open the Windows Command Prompt and change directory to the folder where the file
Rcmd.exe is sitting. This can be done via a command like this:
cd C:\Program Files\R\R-xxx\bin
Step 6 Once there, pass the command
44
Rcmd INSTALL --build johnvanderbrook
By this a (binary) installer broekjan_0.9.zip will be created. This file can be used to
install the library.
B.1.3
How to use the newborn installer
The file broekjan_0.9.zip that was built in the previous subsection can be used to install the
library broekjan. This can be done as follows:
Step 1 Start R.
Step 2 Follow the menu path
Packages −→ Install package(s) from local zip files...
Step 3 Browse to the place where broekjan_0.9.zip resides and double-click the file. The
installation should start then.
Step 4 After installation the library can be loaded into R by passing (under R) the command
library(broekjan)
B.2
Building an R-library under Unix-systems
To build an R-library under Unix-systems (for example Linux, FreeBSD), carry out the following
steps:
Step 1 Load the stuff for your library into the R-workspace.
Step 2 Pass (under R) the command
package.skeleton(name="johnvanderbrook", list=ls())
By this action a folder, named johnvanderbrook, is created in the working directory of
R. This folder is called the skeleton of the library that is to be built.
Step 3 The skeleton contains the files DESCRIPTION, read-and-delete-me and the subfolders data, man, R and src. The file DESCRIPTION must be edited by means of a text
editor, for example like this
Package: broekjan
Type: Package
Title: What the package does (short line)
Version: 0.9
Date: 2008-05-12
Author: Wiebe Pestman
Maintainer: Who to complain to <[email protected]>
Description: More info (maybe more than one line)
License: GNU
The files in the subfolders data and man must also be edited. They contain files with names
blabla.Rd. These files are meant to build the help functions and they can be edited as if
they were simplified LATEX files. Knowledge of LATEX is not required in this.
45
Step 4 If your skeleton is ready, open a console, change directory to the place where your skeleton
resides and pass the command
R CMD build johnvanderbrook
By this a tarball broekjan_0.9.tar.gz will be created. This tarball can be used to
install your library under most Unix systems. Unfortunately not under MacOSX.
Step 5 To install your library, run (as root) the next command in a console
R CMD INSTALL /complete/path/to/broekjan_0.9.tar.gz
By this the library will be installed.
B.3
Building an R-library under MacOSX
Step 1 Load the stuff for your library into the R-workspace.
Step 2 Pass (under R) the command
package.skeleton(name="johnvanderbrook", list=ls())
By this action a folder, named johnvanderbrook, is created in the working directory of
R. This folder is called the skeleton of the library that is to be built.
Step 3 The skeleton contains the files DESCRIPTION, read-and-delete-me and the subfolders data, man, R and src. The file DESCRIPTION must be edited by means of a text
editor, for example like this
Package: broekjan
Type: Package
Title: What the package does (short line)
Version: 0.9
Date: 2008-05-12
Author: Wiebe Pestman
Maintainer: Who to complain to <[email protected]>
Description: More info (maybe more than one line)
License: GNU
The files in the subfolders data and man must also be edited. They contain files with
names of type blabla.Rd. These files are meant to build the help functions and they can
be edited as if they were simplified LATEX files. Knowledge of LATEX is not required in this.
Step 4 If your skeleton is ready, open a console, change directory to the place where your skeleton
resides and pass the command
R CMD build -binary johnvanderbrook
By this a tarball broekjan_0.9.tgz will be created. This tarball can be used to install
the library.
Step 5 To install your library, run (as root) the next command in a console
R CMD INSTALL /complete/path/to/broekjan_0.9.tgz
By this the library will be installed, at least, that is what I hope. Never done this myself...
46
References
[1] DALGAARD, Peter, Introductory Statistics with R, Springer Verlag, Berlin
47
Index
abline( ), 18
ANOVA
1-way, 26
2-way, 27
interaction, 28
appendices, 42
array, 7
atoms, 4
bar plots, 2
Bartlett’s test, 27
binary files, 9, 11
binom.test( ), 32
binomial test, 32, 33
boxplot( ), 15
boxplots, 15
built-in data, 11
characters, 4
chi-square test
on goodness of fit, 34
on independence, 35
closing
a bitstream, 3
a channel, 13
comparing
fractions, see binomial test
Kaplan-Meier curves, see logrank test
means, see ANOVA
medians, see Kruskal-Wallis test
percentages, see binomial test
proportions, see binomial test
two means, see t-test
two medians, see Wilcoxon’s test
two variances, see Fisher’s variance test
variances, see Bartlett’s and Levene’s test
compilers, 44
compound proportions, 33
confidence intervals, 30
confint( ), 30
constant, 30, 31
contingency tables, 33, 35
cos( ), 15
cross tables, 33, 35
cumulative frequency plot, 3
curve( ), 17
data
Excel, 13
exporting, 9
files, 9–11
foreign, 12
importing, 12
SAS, 12
Splus, 12
SPSS, 12
data frames, 8
deinstall, 43
deleting objects, 1
dev.copy( ), 3
dev.off( ), 3
digits, 4
dnorm( ), 17
effect size, 41
exact p-values, 28, 29, 33, 36
Excel data files, 13
exp( ), 15
exporting data, 9
extensions to R, see library
factor( ), 38
factors, 8
files
ascii, 9, 16
binary, 9
native format, 9
plain text, 9, 16
Fisher’s exact test, 36
Fisher’s variance test, 25, 27
fisher.test( ), 36
fitted values, 30
fitted( ), 30
foreign data, 12
frequency plots, 3
frequency tables, 13
functions, 15
Gaussian curves, 1, 17
general linear model, 38
generating random numbers, 14
getwd( ), 9
glm( ), 38
grouping variable, 24
head, 28
48
head( ), 10, 11
histograms, 1, 12, 17
homogeneity of variance
Bartlett’s test, 27
Fisher’s test, 25, 27
Levene’s test, 27
homoscedasticity, 37
horizontal lines, 18
importing foreign data, 13
installation
R-base, 42
R-libraries, 42
interaction, see ANOVA
intercept, 30, 31
interval
confidence, 30
estimate, 30
Kaplan-Meier plot, 39
Kolmogorov-Smirnov
distance, 19
plot, 18
test, 19, 24
Kruskal-Wallis test, 29
KS-distance, 19
KS-plot, 18
KS-test, 24
ks.test( ), 19
labels, 1
Levene’s test, 27, 28
libraries
building, 44
dependencies, 43
install, 42
load, 27, 43
remove, 43
uninstall, 43
likelihood, 38
line
plotting, 18
regression, 31
segments, 31
linear algebra, 6
linear model, 30
linear regression, 30
lines, 18
list, 26
lists, 7
lm( ), 30
loading
data, 11
functions, 16
libraries, 12, 13
objects, 11
log( ), 15
logLik( ), 38
logrank test, 39
ls( ), 1
making functions, 15
Mann-Whitney test, 29
matrix
addition, 6
columns, 6
multiplication, 6
rows, 6
transposition, 6
McNemar’s test, 33
mcnemar.test( ), 33
mean, 14
mean( ), 14
median, 14
median( ), 14
model reduction, 38
native file format, 9, 11
nls( ), 36
nonlinear models, 36
normal curves, 1, 17
numbers, 4
object
array, 7
data frame, 8
factor, 8
list, 7
matrix, 6
vector, 5
objects, 1, 4
percentiles, 15
pin diagram, 2
plain text files, 9, 16
plot
regression, 31
residual, 37
survival, 39
plot( ), 2
plotting curves, 17
49
plotting straight lines, 18
pnorm( ), 22
power, 39
power.anova.test( ), 40
power.prop.test( ), 40
power.t.test( ), 40, 41
predict( ), 30, 38
predictions, 30, 38
probability
cumulative, 21
densities, 21
distributions, 21
prop.test( ), 33
proportion
compound, 33
test, 33
pwr, 40
pwr.2p2n.test( ), 41
pwr.t.test( ), 41
qnorm( ), 22
qq-plots, 20
qqnorm( ), 20
quantile( ), 15
quantiles, 15
quitting a help screen, 12
R-base, 42
R-libraries, 42
random data, 14
random numbers, 14
read.sas( ), 12
read.spss( ), 12
read.table( ), 9
reading in
data, 9
source code, 16
syntax code, 16
regression
conditions, 32
data, 32
generating data, 32
line, 31
linear, 30, 31
logistic, 38
multiple, 31
nonlinear, 36
plots, 31
predictions, 30
through origin, 30, 31
remove.packages( ), 43
removing objects, 1
rep( ), 13
residual analysis, 37
residual scatter plot, 37
residuals( ), 37
rm( ), 1
rnorm( ), 14, 22
RODBC, 13
root, 14, 42, 43
runs test, 29
runs.test( ), 29
sample size, 39
SAS data files, 12
save( ), 11, 16
saving
data, 11
functions, 16
graphics, 3
objects, 11
syntax, 16
sd( ), 14
segments, 31
Shapiro-Wilk test, 37
shapiro.test( ), 37
sin( ), 15
source( ), 16
split( ), 26
splitting datasets, 26
SPSS data files, 12
sqrt( ), 14
square root, 14
squaring numbers, 15
standard deviation, 14
step( ), 38
straight lines, 18
strings, 4
subset( ), 26
summary, 30, 38
summary( ), 30, 38
Surv( ), 39
survdiff( ), 39
survfit( ), 39
survival analysis, 39
syntax, 16
t( ), 6
t-test
1-sample, 23
50
2-sample, 24
paired-sample, 24
t.test( ), 23
table( ), 13
tables, 33
tabulated data, 13
tail, 28
tail( ), 10, 11
transforming data, 15
trimmed mean, 14
user-defined functions, 15
variance, 14, 25
variance test
Bartlett’s test, 27
Fisher’s test, 25, 27
Levene’s test, 27
variance( ), 14
vector, 5
vertical lines, 18
Wilcoxon, see also Mann-Whitney
rank-sum test, 29
signed-rank test, 28
test, 28
workspace, 1
write( ), 9
51